Cornelis Networks, a company spun off from Intel, has closed a $205 million funding round led by IAG Capital Partners and simultaneously unveiled Active Compute Fabric, an open-architecture, GPU-agnostic networking layer designed to compete directly with Nvidia's dominant InfiniBand and NVLink interconnect technologies. Alongside the funding announcement, the company revealed a strategic collaboration with Qualcomm focused on rack-scale AI inference infrastructure.
The announcement lands at a pivotal moment for the AI infrastructure market, where Nvidia's networking technologies have become deeply entrenched as the default interconnect standard for large-scale AI training and inference clusters, giving the company substantial pricing power and strategic control over how hyperscalers and enterprises architect their AI computing infrastructure. Cornelis Networks' pitch — an open, GPU-agnostic alternative — speaks directly to a growing appetite among large AI infrastructure buyers for interconnect solutions that reduce vendor lock-in and preserve flexibility to mix and match chip suppliers as the broader AI hardware market diversifies beyond Nvidia's historical dominance.
Cornelis Networks traces its lineage to Intel's high-performance computing interconnect business, giving the company deep institutional expertise in networking technology originally developed for supercomputing and scientific computing applications, now being repositioned for the commercial AI infrastructure market. That heritage provides Cornelis with a credible technical foundation as it seeks to position Active Compute Fabric as a genuine alternative to Nvidia's ecosystem, rather than simply a lower-cost commodity substitute.
The strategic collaboration with Qualcomm on rack-scale AI inference infrastructure adds further weight to Cornelis's positioning, pairing its networking technology with a major semiconductor player that has itself been expanding its ambitions in AI inference chips as an alternative to Nvidia's graphics processing units for certain workloads. Rack-scale inference infrastructure — the systems architecture used to deploy trained AI models for real-world use at scale — has become an increasingly important battleground within the AI hardware ecosystem, as the industry's overall compute demand shifts gradually from a training-dominated phase toward one where inference workloads represent a growing and, in some projections, eventually larger share of total AI infrastructure spending.
The $205 million funding round underscores continued investor appetite for companies positioned to challenge Nvidia's dominant position within AI networking and interconnect infrastructure, even as questions persist across the industry about how quickly genuine alternatives can achieve the scale, reliability and software ecosystem maturity needed to meaningfully dent Nvidia's market share among the largest AI infrastructure buyers, including major cloud hyperscalers and frontier AI labs.
For enterprises and hyperscalers building out AI infrastructure, the emergence of credible open, GPU-agnostic networking alternatives such as Active Compute Fabric could, over time, meaningfully reshape the economics and architectural flexibility of large-scale AI deployments, particularly if the technology can demonstrate performance and reliability on par with established Nvidia interconnect standards at competitive pricing.
